I have managed to compile the module and the tools on 18.04.
It needs to add some headers , change version of opencv and it compile without errors.
ODM is 0.9.1
calibrate_video.py is working , but I can`t see those options in args :
/code/run.py -help
–video ## Path to the video file to process
–slam-config ## Path to config file for orb-slam
Do you know how to expose them ?
i would like to test odm_slam feature?
Real-time ?
Hey, from the Git history, I could see that SLAM module is quite outdated(last commit was 3 years ago!). SLAM based mapping is gaining quite some popularity for the past few months. I’d love to revamp the whole SLAM module. Before that, I’d love to hear the thoughts of the community!!
Yup. Users can have the option to choose this particular method for the images they feed to ODM. On a side note, can SLAM based processing compete with the one in the existing pipeline of ODM, in terms of accuracy & quality??
As a rule, SLAM tends to have better localised accuracy and worse global accuracy. But, as you may know better than I: implementation matters a lot and there are clever ways to perform global and semi-global corrections.
Here’s a relatively new paper, published in 2020. Results are really fast and accuracy comparable with Pix4D. Although there are many works out there, like TerrainFusion, Map2DFusion, this is even faster than the ones mentioned here. Can you have a look at this? I really want to get started on adding SLAM to ODM pipeline. Meanwhile, I’ll do some digging in ODM’s existing SFM pipeline, and how SLAM can be added to the existing design.